Summary
Bennett Fenner of Zed walks through a live coding session building an ACP (Agent Client Protocol) compatible coding agent in TypeScript, demonstrating the open standard Zed has developed to let any AI coding agent work inside any compatible editor through a unified interface. ACP is a JSON-RPC protocol — conceptually similar to MCP (Model Context Protocol) or LSP (Language Server Protocol) — designed so that agents and clients can communicate without bespoke per-pairing integrations.
The motivation came from the proliferation of terminal-based coding agents in 2025: Claude Code, OpenAI Codex, Gemini CLI, and others each required separate integration work for editor tooling. Zed’s answer was a protocol layer. At the time of the talk, ACP has around 40 client implementations including JetBrains, Obsidian, and Cursor (which has native ACP mode built into its CLI agent), plus adapters for agents that want to expose their existing interfaces through ACP.
The live demo starts from a minimal coding agent with two tools (read file, edit file) and adds ACP support incrementally: session lifecycle management, streaming token output via session-update notifications wired to Anthropic’s TypeScript SDK streaming API, tool call events that surface in Zed’s built-in ACP debug panel, and cancellation support. The result is a functioning ACP agent that streams responses and exposes tool calls inside Zed — built in roughly ten minutes from a TypeScript boilerplate. Fenner notes that registering the agent in Zed requires only a single config line pointing to the compiled agent script.
📺 Source: AI Engineer · Published July 08, 2026
🏷️ Format: Tutorial Demo
